Amazon.com Inc.

09/18/2025 | Press release | Distributed by Public on 09/18/2025 16:23

Qwen3 and DeepSeek-V3.1 models now available fully managed on AWS

Key takeaways

  • The Qwen3 family includes model options for sophisticated coding and general reasoning, while DeepSeek-V3.1 demonstrates exceptional performance across math, coding, and agentic tasks.
  • Unlike proprietary models, open weight models offer developers transparency into model weights, and as a result, provide customers with enhanced opportunities for customization.
  • The announcement expands the existing broad selection of fully managed open weight models in Amazon Bedrock, including those from Meta, Mistral AI, and most recently OpenAI.
Amazon Web Servicesis expanding the already broad lineup of fully managed, industry-leading models in Amazon Bedrockwith the addition of open weight Qwen3 from Alibaba and DeepSeek-V3.1 models, which are now available globally.
The addition of these models enhances the existing selection of open weight models in Amazon Bedrock from providers like Meta, Mistral AI, and OpenAI. This news further underscores AWS's commitment to offering customers a broad selectionof cutting-edge AI technologies, so they can select the best models for their use case, while benefiting from AWS's security, privacy, and reliability.
By using these models in Amazon Bedrock, customers benefit from enterprise-grade security, including data encryption and strict access controls, which help customers maintain data privacy and regulatory compliance. Customers retain full control over their data, meaning AWS does not share their model input and output data with model providers, and it is not used to improve the base models. Additionally, customers can set up safeguards such as Amazon Bedrock Guardrails, which are recommended by AWS to detect and prevent hallucinations.

Meet Project Rainier, Amazon's one-of-a-kind machine ushering in the next generation of AI

Our new AI "compute cluster" will connect hundreds of thousands of Trainium2 chips across the U.S.

Why you should care

AWS believes that giving customers access to a diverse range of both proprietary and open weight AI models is crucial for unlocking global customers' freedom to innovatewith generative AI. Open weight models allow developers more visibility into their 'recipe', helping to enhance customers' ability to customize them and fostering innovation, as well as positively impacting price performance.
Within the Qwen3 family, AWS customers now have access to four new open weight models. These models can plan multistep workflows, integrate with tools and APIs, and handle long context windows within a single task, and the two general purpose models provide both 'thinking' and 'non-thinking' reasoning modes.
  • Qwen3 models are multilingual, with particularly strong performance in Chinese and English, enabling seamless cross-cultural business operations and content generation.
  • Qwen3-Coder-480B-A3B-Instruct and Qwen3-Coder-30B-A3B are optimized for complex software engineering scenarios, including code generation and understanding and advanced agentic tasks. They are both capable of writing code in many programming languages and using other digital tools autonomously (i.e., external tools and applications).
  • Qwen3-235B-A22B-Instruct-2507 is designed for general purpose reasoning, balancing capability with efficiency, and delivering competitive performance across coding, math, and general reasoning tasks. 'Mixture of expert' (MoE) models like these activate only part of their parameters for each request, meaning they only use part of their knowledge for a specific task or question, resulting in high performance and efficiency.
  • Qwen3-32B (Dense) is appropriate for computational tasks and applications with limited computational resources, or scenarios that require consistent and predictable performance. 'Dense' models are usually smaller and designed to have all parts work together all the time, unlike MoE models, which work more like specialized departments, activating only when their specific expertise is needed.
The latest DeepSeekmodel, DeepSeek-V3.1, offers hybrid reasoning capabilities that balance fast responses with deep, transparent thinking.
  • Customers can toggle between either a 'thinking mode' that carefully works through problems step-by-step or a 'quick response mode' for more straightforward questions and have visibility into the model's decision-making process.
  • The highly sophisticated model is comparable to some of the most advanced AI systems available today, yet its MoE architecture means that customers can benefit from its performance while optimizing compute costs.
  • DeepSeek-V3.1 is multilingual and particularly adept at software development, mathematical reasoning, and data analysis, delivering superior performance across coding and technical challenges. It is also well suited for agentic problem-solving tasks like building AI agents and process automation.

A city in the palm of your hand: Exploring the intricate world of an Amazon Web Services chip

Tour the micro metropolis where calculations run 24/7 and data commutes at light speed.

Straight from the source, an AWS service team leader on open models

According to Luis Wang, director, Amazon Bedrock, "Open weight models represent a vital frontier in AI innovation, which is why we have invested in making AWS the best place to run them securely, at scale, and cost effectively. We don't think there's going to be one model that's the best model for every use case…Many customers love using open models, and one of the benefits of an open model is you have a lot more flexibility to go play with it."

Meet the AI: Qwen3 Family

If the Qwen3 models were people, the Qwen3-Coder models would be your detail-oriented friends who can follow complicated furniture assembly instructions without getting frustrated, turning a jumble of parts into a perfectly built bookcase while explaining each step clearly. The more general-purpose Qwen3 models are like an impressively multilingual family. They speak dozens of languages fluently and share an encyclopedic knowledge of diverse subjects, from explaining scientific concepts to writing creative stories. They can have thoughtful conversations about almost anything, remembering every detail from your previous chats, no matter how long ago they were.

Meet the AI: DeepSeek-V3.1

If DeepSeek-V3.1 was a person, it would be your problem-solving friend who systematically breaks down challenges through logical reasoning while adapting their approach based on the complexity of your question.

Crunching the numbers

  • The Qwen3-Coder-30B-A3B-Instruct and Qwen3-235B-A22B-Instruct-2507 models can process up to 262K tokens at once. That's equivalent to about 200,000 words or roughly two full-length novels in a single conversation.
  • DeepSeek-V3.1 contains 685 billion parameters. Parameters are a model's 'knowledge connections', internal numerical settings that are adjusted during training to help it learn from data and make predictions. DeepSeek-V3.1 only uses the relevant portions of its knowledge for each task, making it more cost-efficient to run while still delivering powerful performance.
  • According to Artificial Analysis, DeepSeek-V3.1 demonstrates significant improvements over its predecessors, most notably in reasoning and agentic skills.
  • Customers like CoinMarketCap, a source for real-time cryptocurrency market data, are benefitting from DeepSeek's models in Amazon Bedrock and have been using DeepSeek-R1 to transform large volumes of crypto data from the site into insights for its users.

The bigger story

Open weight models represent a significant advancement in transparent and flexible AI. You can think of them similar to a soup base where, while the existing flavors are often delicious as-is, chefs can build upon the recipe with additional ingredients that add to the depth and complexity of the resulting soup. Similarly, customers can review the recipe, adjust according to their tastes, or even build upon it with their own innovations. As such, open weight models are helping to reshape generative AI's future in several important ways:
  • They foster innovation by enabling customers to build upon existing work rather than starting from scratch.
  • The increased understanding around an AI's learned knowledge helps to improve transparency for customers.
  • When delivered as fully managed models in Amazon Bedrock, they are accessible without requiring specialized technical teams.

Your guide to free and low-cost AWS courses that can help you use generative AI

More than 135 AWS trainings on AI/ML are available to everyone, with all levels of experience.

What's around the corner

AWS believes that giving customers access to diverse AI models is crucial for unlocking the full potential of generative AI. For example, not long ago we introduced OpenAI's open weight models to Amazon Bedrock, and today we are expanding availability of those models to every continent with an AWS Region, spanning US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), Asia Pacific (Mumbai), Europe (Ireland), Europe (London), Europe (Milan), Europe (Stockholm), and South America (São Paolo).
As we make more of these powerful open weight models accessible through services like Amazon Bedrock, organizations without specialized AI expertise can implement solutions that were previously available only to those with significant technical resources and infrastructure. This in turn could accelerate innovation across healthcare, financial services, manufacturing, and countless other sectors, enabling more sophisticated automated workflows, scientific research assistance, complex mathematical analysis, and business process optimization.

Dive deeper

Visit the AWS News blog for more information on Qwen3and DeepSeek-V3.1models.

Ready to explore?

Get started with these open weight models in the Amazon Bedrock console.

Trending news and stories

  1. Prime Big Deal Days is back October 7-8: Here are 5 shopping tips to save big during the event
  2. Prime Big Deal Days 2025: 8 of the best early deals you can shop right now
  3. Here's how much a Prime membership costs, and how to make the most of its benefits in time for Prime Big Deal Days this October
  4. 5 awesome benefits that make your Prime membership worth it

Related Tags

  1. AWS
  2. Artificial Intelligence
  3. Technology
  4. Innovation
Share
Amazon.com Inc. published this content on September 18, 2025, and is solely responsible for the information contained herein. Distributed via Public Technologies (PUBT), unedited and unaltered, on September 18, 2025 at 22:23 UTC. If you believe the information included in the content is inaccurate or outdated and requires editing or removal, please contact us at [email protected]